CBP Labels AI Model Cards as ‘Law Enforcement Privileged’
Customs and Border Protection (CBP) has withheld significant details from model cards describing two AI-based border security tools. These cards, which should clarify how the AI systems work and their intended uses, were heavily redacted when released under a Freedom of Information Act (FOIA) request. The agency cited concerns that full disclosure could compromise law enforcement techniques.
What Are Model Cards?
Model cards serve as transparent documents that outline the design, function, and potential risks of AI systems. They help explain the technical makeup and operational context of algorithms. According to experts, such as Varoon Mathur, a former AI engineer with government experience, these cards are common tools to communicate engineering requirements and limitations.
However, the sensitive nature of some information often leads agencies to limit public access. CBP’s approach reflects this dilemma, balancing transparency with operational security.
Details of the Redacted AI Systems
- Passport Anomaly Model: Designed to help officers identify passports that need further scrutiny. It analyzes indicators like passport number, issue date, date of birth, and issuing country.
- Commodity Detection Model: Uses computer vision, neural networks, and object detection to assign commodity codes to images.
CBP redacted nearly all technical and performance details of both models. The agency explained that revealing training data, validation methods, or system weaknesses could expose law enforcement methods and enable adversaries to evade detection.
Interestingly, one page from the passport anomaly model was partially disclosed, showing it employs an ensemble model combining logistic regression, conditional probability, date ranges, and association rules. It uses data from various sources, including the Automated Targeting System-Passenger (ATS-P) and electronic chip data.
Legal and Transparency Concerns
Advocates for government transparency question CBP’s broad use of law enforcement exemptions under FOIA. Alex Howard, a transparency expert, noted that these redactions might be overly sweeping and subject to challenge. He emphasized the public interest in knowing how AI tools are used in law enforcement, including access to training data to assess potential biases or discrimination.
Other agencies, such as the State Department and Federal Aviation Administration, have released more detailed AI system information in similar contexts. This contrast raises questions about consistency in transparency practices across government bodies.
Implications for Legal Professionals
For those in legal roles, these developments underline the tension between operational secrecy and the need for accountability in AI-driven law enforcement tools. Understanding the scope of redactions and the justifications for withholding information is crucial when evaluating compliance with transparency laws and assessing potential impacts on civil rights.
Legal professionals should remain informed about the evolving standards for AI transparency in government systems and consider how FOIA exemptions are applied in practice.
For further resources on AI applications and governance, explore training offerings at Complete AI Training.